Re: [PATCH v3 5/7] x86/dmop: Add XEN_DMOP_{bind,unbind}_pt_msi_irq DM ops

[Jan Beulich <jbeulich@xxxxxxxx>]

On 11.03.2026 17:04, Jan Beulich wrote:
> On 09.03.2026 13:31, Julian Vetter wrote:
>> @@ -607,6 +631,83 @@ int dm_op(const struct dmop_args *op_args)
>>          break;
>>      }
>>  
>> +    case XEN_DMOP_bind_pt_msi_irq:
>> +    {
>> +        const struct xen_dm_op_bind_pt_msi_irq *data =
>> +            &op.u.bind_pt_msi_irq;
>> +        struct xen_domctl_bind_pt_irq bind = {
>> +            .machine_irq = data->machine_irq,
>> +            .irq_type = PT_IRQ_TYPE_MSI,
>> +        };
>> +        int irq;
>> +
>> +        rc = -EINVAL;
>> +        if ( data->pad0 || data->pad1 )
>> +            break;
>> +
>> +        if ( data->flags & ~XEN_DMOP_MSI_FLAG_UNMASKED )
>> +            break;
>> +
>> +        irq = domain_pirq_to_irq(d, bind.machine_irq);
>> +
>> +        rc = -EPERM;
>> +        if ( irq <= 0 || !irq_access_permitted(current->domain, irq) )
>> +            break;
>> +
>> +        bind.u.msi.gvec = data->gvec;
>> +        bind.u.msi.gflags =
>> +            msi_addr_to_gflags(data->addr, data->data,
>> +                               !(data->flags & XEN_DMOP_MSI_FLAG_UNMASKED));
>> +        bind.u.msi.gtable = data->gtable;
>> +
>> +        rc = -ESRCH;
>> +        if ( is_iommu_enabled(d) )
>> +        {
>> +            pcidevs_lock();
>> +            rc = pt_irq_create_bind(d, &bind);
>> +            pcidevs_unlock();
> 
> I understand the same locking is used at the other call site, but it's as
> questionable there as it is here. We should try hard to avoid use of this
> global lock when something lighter-weight would do.

I should perhaps have added that I have this on my own list of things to look
into, just that I didn't get to it yet.

Jan